## Finding Almost Squares II

### November 27, 2006

Abstract
In this article, we study short intervals that contain another type of “almost square”, an integer $n$  which can be factored in two different ways $n={a}_{1}{b}_{1}={a}_{2}{b}_{2}$  with ${a}_{1},{a}_{2},{b}_{1},{b}_{2}$  close to $\sqrt{n}$  .

1 Introduction

In [ 1 ], the author studied the problem of finding “almost squares” in short intervals, namely:
Question 1. For $0\le \theta <1/2$  , what is the least $f\left(\theta \right)$  such that, for some ${c}_{1},{c}_{2}>0$  , any interval $\left[x-{c}_{1}{x}^{f\left(\theta \right)},x+{c}_{1}{x}^{f\left(\theta \right)}\right]$  contains an integer $n$  with $n=ab$  , where $a$  , $b$  are integers in the interval $\left[{x}^{1/2}-{c}_{2}{x}^{\theta },{x}^{1/2}+{c}_{2}{x}^{\theta }\right]$  ? Note: ${c}_{1}$  and ${c}_{2}$  may depend on $\theta$  .
A similar question is the following:
Question 2. For $0\le \theta <1/2$  , what is the least $g\left(\theta \right)$  such that, for some ${c}_{1},{c}_{2}>0$  , any interval $\left[x-{c}_{1}{x}^{g\left(\theta \right)},x+{c}_{1}{x}^{g\left(\theta \right)}\right]$  contains an integer $n$  with $n={a}_{1}{b}_{1}={a}_{2}{b}_{2}$  , where ${a}_{1}<{a}_{2}\le {b}_{2}<{b}_{1}$  are integers in the interval $\left[{x}^{1/2}-{c}_{2}{x}^{\theta },{x}^{1/2}+{c}_{2}{x}^{\theta }\right]$  ? Note: ${c}_{1}$  and ${c}_{2}$  may depend on $\theta$  .
Note: Actually, the author first considered Question  2 and then turned to Question  1 which has connection to the problem on the distribution of ${n}^{2}\alpha \left(mod1\right)$  and the problem on gaps between sums of two squares.
In [ 1 ], we showed that $f\left(\theta \right)=1/2$  when $0\le \theta <1/4$  , $f\left(1/4\right)=1/4$  and $f\left(\theta \right)\ge 1/2-\theta$  . We conjectured that $f\left(\theta \right)=1/2-\theta$  for $1/4<\theta <1/2$  and gave conditional result when $1/4<\theta <3/10$  . To Question  2 , we have the following
Theorem 1.1. For $0<\theta <1/4$  , $g\left(\theta \right)$  does not exist (i.e. all possible products of pairs of integers in $\left[{x}^{1/2}-{c}_{2}{x}^{\theta },{x}^{1/2}+{c}_{2}{x}^{\theta }\right]$  are necessarily distinct for large $x$  ).
Theorem 1.2. For $1/4\le \theta <1/2$  , $g\left(\theta \right)\ge 1-2\theta$  .
Theorem 1.3. For $1/4\le \theta \le 1/3$  , $g\left(\theta \right)\le 1-\theta$  .
We believe that the lower bound is closer to the truth and conjecture
Conjecture 1.1. For $1/4\le \theta <1/2$  , $g\left(\theta \right)=1-2\theta$  .

2 Preliminaries and $0\le \theta <1/4$

Suppose $n={a}_{1}{b}_{1}={a}_{2}{b}_{2}$  with ${x}^{1/2}-{c}_{2}{x}^{\theta }\le {a}_{1}<{a}_{2}\le {b}_{2}<{b}_{1}\le {x}^{1/2}+{c}_{2}{x}^{\theta }$  .
Let ${d}_{1}=\left({a}_{1},{a}_{2}\right)$  and ${d}_{2}=\left({b}_{1},{b}_{2}\right)$  be the greatest common divisors. Then we must have ${d}_{1},{d}_{2}>1$  . For otherwise, say ${d}_{1}=1$  , then ${a}_{2}$  divides ${b}_{1}$  which implies ${x}^{1/2}+{c}_{2}{x}^{\theta }\ge {b}_{1}\ge 2{a}_{2}\ge 2{x}^{1/2}-2{c}_{2}{x}^{\theta }$  . This is impossible for large $x$  as $\theta <1/2$  . Now, let ${a}_{1}={d}_{1}{e}_{1}$  , ${a}_{2}={d}_{1}{e}_{2}$  , ${b}_{1}={d}_{2}{f}_{1}$  and ${b}_{2}={d}_{2}{f}_{2}$  . Here $\left({e}_{1},{e}_{2}\right)=1=\left({f}_{1},{f}_{2}\right)$  . Then $n={d}_{1}{e}_{1}{d}_{2}{f}_{1}={d}_{1}{e}_{2}{d}_{2}{f}_{2}\text{gives}{e}_{1}{f}_{1}={e}_{2}{f}_{2}.$  Due to co-primality, ${e}_{2}={f}_{1}$  and ${e}_{1}={f}_{2}$  . Therefore,
 $\begin{array}{c}n=\left({d}_{1}{e}_{1}\right)\left({d}_{2}{e}_{2}\right)=\left({d}_{1}{e}_{2}\right)\left({d}_{2}{e}_{1}\right)\end{array}$ (1)
with $1<{d}_{1}<{d}_{2}$  , ${e}_{1}<{e}_{2}$  and $\left({e}_{1},{e}_{2}\right)=1$  .
Now, from ${a}_{2}-{a}_{1}\le 2{c}_{2}{x}^{\theta }$  , ${d}_{1}\le {d}_{1}{e}_{2}-{d}_{1}{e}_{1}\le 2{c}_{2}{x}^{\theta }$  . Similarly, one can deduce that ${d}_{2},{e}_{1},{e}_{2}\le 2{c}_{2}{x}^{\theta }$  . Moreover, as ${d}_{1}{e}_{1}={a}_{1}\ge {x}^{1/2}-{c}_{2}{x}^{\theta }$  , we have ${d}_{1},{e}_{1}\ge \frac{1}{2{c}_{2}}{x}^{1/2-\theta }-\frac{1}{2}$  . Similarly, ${d}_{2},{e}_{2}\ge \frac{1}{2{c}_{2}}{x}^{1/2-\theta }-\frac{1}{2}$  . Summing up, we have
 $\begin{array}{c}\frac{1}{2{c}_{2}}{x}^{1/2-\theta }-\frac{1}{2}\le {d}_{1},{d}_{2},{e}_{1},{e}_{2}\le 2{c}_{2}{x}^{\theta }.\end{array}$ (2)
From ( 2 ), we see that no such $n$  exists for $0\le \theta <1/4$  and hence Theorem  1.1 .

3 Lower bound for $g\left(\theta \right)$

From ( 1 ) and ( 2 ), we see that an integer $n={a}_{1}{b}_{1}={a}_{2}{b}_{2}$  , satisfying the conditions for ${a}_{1},{a}_{2},{b}_{1},{b}_{2}$  in Question  2 , must be of the form:
$n=\left({d}_{1}{e}_{1}\right)\left({d}_{2}{e}_{2}\right)\text{with}\frac{1}{2{c}_{2}}{x}^{1/2-\theta }-\frac{1}{2}\le {d}_{1},{d}_{2},{e}_{1},{e}_{2}\le 2{c}_{2}{x}^{\theta }$  and ${x}^{1/2}-{c}_{2}{x}^{\theta }\le {d}_{1}{e}_{1}<{d}_{1}{e}_{2},{d}_{2}{e}_{1}<{d}_{2}{e}_{2}\le {x}^{1/2}+{c}_{2}{x}^{\theta }$  . In particular, ${e}_{2}{d}_{2}-{e}_{2}{d}_{1}\le 2{c}_{2}{x}^{\theta }$  which implies ${e}_{2}-{e}_{1}\le 2{c}_{2}{x}^{\theta }/{d}_{2}$  . Similarly, ${d}_{2}-{d}_{1}\le 2{c}_{2}{x}^{\theta }/{e}_{2}$  . Thus, the number of such quartuple $\left({d}_{1},{d}_{2},{e}_{1},{e}_{2}\right)$  is bounded by $\ll {{\sum }_{{x}^{1/2-\theta }\ll {d}_{2},{e}_{2}\ll {x}^{\theta }}}_{{x}^{1/2}-{c}_{2}{x}^{\theta }\le {d}_{2}{e}_{2}\le {x}^{1/2}+{c}_{2}{x}^{\theta }}\frac{{x}^{\theta }}{{e}_{2}}\frac{{x}^{\theta }}{{d}_{2}}\ll \frac{{x}^{2\theta }}{{x}^{1/2}}{x}^{\theta }{x}^{\epsilon }={x}^{3\theta -1/2+\epsilon }$  for any $\epsilon >0$  as the number of divisor function $d\left(n\right)\ll {n}^{\epsilon }$  . It follows that there are at most $O\left({x}^{3\theta -1/2+\epsilon }\right)$  such integers $n$  in the interval $\left[x-{c}_{2}{x}^{1/2+\theta }/3,x+{c}_{2}{x}^{1/2+\theta }/3\right]$  . Therefore, some two consecutive such $n$  's have gap $\gg \frac{{x}^{1/2+\theta }}{{x}^{3\theta -1/2+\epsilon }}={x}^{1-2\theta -\epsilon }.$  Pick $y$  to be the midpoint between these two integers. Then, for some constant $c>0$  , the interval $\left[y-c{y}^{1-2\theta -\epsilon },y+c{y}^{1-2\theta -\epsilon }\right]$  does not contain any integer $n={a}_{1}{b}_{1}={a}_{2}{b}_{2}$  with ${y}^{1/2}-{c}_{2}{y}^{\theta }/2\le {a}_{1}<{a}_{2}\le {b}_{2}<{b}_{1}\le {y}^{1/2}+{c}_{2}{y}^{\theta }/2$  as $x-{c}_{2}{x}^{1/2+\theta }/3\le y\le x+{c}_{2}{x}^{1/2+\theta }/3$  . Consequently, for any constant $c,{c}^{\prime }>0$  , there is arbitrarily large $y$  such that the interval $\left[y-c{y}^{1-2\theta -2\epsilon },y+c{y}^{1-2\theta -2\epsilon }\right]$  does not contain any integer $n={a}_{1}{b}_{1}={a}_{2}{b}_{2}$  with ${y}^{1/2}-{c}^{\prime }{y}^{\theta }\le {a}_{1}<{a}_{2}\le {b}_{2}<{b}_{1}\le {y}^{1/2}+{c}^{\prime }{y}^{\theta }$  . Therefore, $g\left(\theta \right)\ge 1-2\theta -2\epsilon$  which gives Theorem  1.2 by letting $\epsilon \to 0$  .

4 Upper bound for $g\left(\theta \right)$

Proof of Theorem  1.3 : For any large $x$  , set $N=\left[{x}^{1/4}\right]$  and $\xi =\left\{{x}^{1/4}\right\}$  , the integer part and fractional part of ${x}^{1/4}$  respectively. Based on ( 1 ), we are going to pick, for $0\le \epsilon \le 1/2$  ,
 $\begin{array}{c}{d}_{1}=qN+{r}_{1},{d}_{2}=qN+{r}_{2},{e}_{1}=\frac{N+{s}_{1}}{q},{e}_{2}=\frac{N+{s}_{2}}{q}\end{array}$ (3)
for some $1\le q\le {N}^{\epsilon }$  , $0\le {r}_{1},{r}_{2}  and ${s}_{1},{s}_{2}\ll q$  with $N\equiv -{s}_{1}\equiv -{s}_{2}\left(modq\right)$  . Our goal is to make $x=\left(N+\xi {\right)}^{4}={N}^{4}+4{N}^{3}\xi +O\left({N}^{2}\right)\approx \left(qN+{r}_{1}\right)\frac{N+{s}_{1}}{q}\left(qN+{r}_{2}\right)\frac{N+{s}_{2}}{q}.$  The right hand side above is
 $\begin{array}{c}\begin{array}{cccccc}=& \left[{N}^{2}+\left(\frac{{r}_{1}}{q}+{s}_{1}\right)N+\frac{{r}_{1}{s}_{1}}{q}\right]\left[{N}^{2}+\left(\frac{{r}_{2}}{q}+{s}_{2}\right)N+\frac{{r}_{2}{s}_{2}}{q}\right]& =& {N}^{4}+\left(\frac{{r}_{1}+{r}_{2}}{q}+{s}_{1}+{s}_{2}\right){N}^{3}+\left[\frac{{r}_{1}{s}_{1}}{q}+\frac{{r}_{2}{s}_{2}}{q}+\left(\frac{{r}_{1}}{q}+{s}_{1}\right)\left(\frac{{r}_{2}}{q}+{s}_{2}\right)\right]{N}^{2}& & +\left[\frac{{r}_{1}{s}_{1}}{q}\left(\frac{{r}_{2}}{q}+{s}_{2}\right)+\frac{{r}_{2}{s}_{2}}{q}\left(\frac{{r}_{1}}{q}+{s}_{1}\right)\right]N+\frac{{r}_{1}{s}_{1}{r}_{2}{s}_{2}}{{q}^{2}}\end{array}\end{array}$ (4)
By Dirichlet's Theorem on diophantine approximation, we can find integer $1\le q\le {N}^{\epsilon }$  such that $|4\xi -\frac{p}{q}|\le \frac{1}{q{N}^{\epsilon }}$  for some integer $p$  . Fix such a $q$  . Then, pick ${s}_{1}<{s}_{2}<0$  to be the largest two integers such that $N\equiv -{s}_{1}\equiv -{s}_{2}\left(modq\right)$  . Clearly, ${s}_{1},{s}_{2}\ll q$  . Then, one simply picks some $0<{r}_{1}<{r}_{2}\ll {q}^{2}$  such that $\frac{{r}_{1}+{r}_{2}}{q}+{s}_{1}+{s}_{2}=\frac{p}{q}$  . With these values for $q,{r}_{1},{r}_{2},{s}_{1},{s}_{2}$  , ( 4 ) is $={N}^{4}+4{N}^{3}\xi +O\left({N}^{3-\epsilon }\right)+O\left({q}^{2}{N}^{2}\right)+O\left({q}^{3}N\right)+O\left({q}^{4}\right).$  Hence, we have just constructed an integer $n={d}_{1}{e}_{1}{d}_{2}{e}_{2}$  which is within $O\left({N}^{3-\epsilon }\right)+O\left({N}^{2+2\epsilon }\right)=O\left({x}^{3/4-\epsilon /4}\right)+O\left({x}^{1/2+\epsilon /2}\right)=O\left({x}^{3/4-\epsilon /4}\right)$  from $x$  if $\epsilon \le 1/3$  . One can easily check that ${a}_{1}={d}_{1}{e}_{1}$  , ${b}_{1}={d}_{2}{e}_{2}$  , ${a}_{2}={d}_{1}{e}_{2}$  and ${b}_{2}={d}_{2}{e}_{1}$  are in the interval $\left[{x}^{1/2}-C{x}^{1/4+\epsilon /4},{x}^{1/2}+C{x}^{1/4+\epsilon /4}\right]$  for some constant $C>0$  . Set $\theta =1/4+\epsilon /4$  . We have, for some ${C}^{\prime }>0$  , $n={a}_{1}{b}_{1}={a}_{2}{b}_{2}$  in the interval $\left[x-{C}^{\prime }{x}^{1-\theta },x+{C}^{\prime }{x}^{1-\theta }\right]$  such that ${a}_{1}<{a}_{2},{b}_{2}<{b}_{1}$  are integers in $\left[{x}^{1/2}-C{x}^{\theta },{x}^{1/2}+C{x}^{\theta }\right]$  provided $1/4\le \theta \le 1/4+1/12=1/3$  . This proves Theorem  1.3 .

5 Open questions

Conjecture  1.1 may be too hard to prove in the moment. As a starting point, can one show that $g\left(1/4\right)=1/2$  ? Or even $g\left(1/4\right)<3/4$  ? Another possibility would be trying to get some results conditionally like [ 1 ]. Also, one may consider $g\left(\theta \right)$  when $\theta$  is near to $1/2$  . This leads to the problem about gaps between integers that have more than one representation as a sum of two squares.
Acknowledgement The author would like to thank the American Institute of Mathematics for providing a stimulating environment to work at.
References

1. Tsz Ho Chan, Finding Almost Squares, preprint, 2005, arXiv:math.NT/0502199.

Tsz Ho Chan American Institute of Mathematics 360 Portage Avenue Palo Alto, CA 94306 U.S.A. thchan@aimath.org